An Interpretable Radiomics-based ML Model for Predicting Reverse Left Ventricular Remodeling in STEMI Patients Using LGE of Myocardial Scar

Motivation: Current imaging techniques face limitations in accurately predicting reverse left ventricular remodeling (r-LVR) in STEMI patients, an essential factor for guiding post-infarction treatment. This gap motivated us to explore advanced imaging and analysis methods to improve prediction accuracy. Goal(s): This study aimed to improve r-LVR prediction by combining LGE-scar radiomics with conventional CMR markers in a LightGBM model. Approach: We integrated radiomic features and CMR markers, using SHAP analysis to interpret the model's predictions. Results: The model achieved high accuracy (AUC: 0.890), surpassing traditional methods. SHAP analysis confirmed the interpretability of selected radiomic features, underscoring their clinical potential in supporting personalized treatment strategies. Impact: This study advances r-LVR prediction in STEMI patients improving clinical outcomes. By combining radiomics with traditional CMR markers, it enables a deeper understanding of myocardial remodeling processes, potentially guiding future studies on cardiac tissue characterization and predictive modeling.